ACL 2025 Papers

The CUTE benchmark showed that LLMs struggle with character understanding in English. We extend it to more languages with diverse scripts and writing systems, introducing EXECUTE. Our simplified framework allows easy expansion to any language. Tests across multiple LLMs reveal that challenges in other languages are not always on the character level as in English. Some languages show word-level processing issues, some show no issues at all. We also examine sub-character tasks in Chinese, Japanese, and Korean to assess LLMs’ understanding of character components.

Event temporal reasoning (ETR) aims to model and reason about the relationships between events and time, as well as between events in the real world. Proficiency in ETR is a significant indicator that a large language model (LLM) truly understands the physical world. Previous question-answering datasets available for evaluating the ETR ability lack a systematic taxonomy and pay limited attention to compound questions. In this paper, we propose a unified taxonomy for event temporal questions and construct a comprehensive benchmark ETRQA, to evaluate the ETR abilities of LLMs based on this taxonomy. ETRQA not only inherits and expands the evaluation content of existing datasets but also contains multiple categories of compound questions. We evaluate two leading LLM series, Llama and Qwen, on ETRQA across various settings. Our experimental results indicate that large-scale LLMs exhibit certain ETR abilities. Yet they do not perform well in solving specific types of reasoning tasks, including reasoning involving time spans, reasoning for compound questions, and reasoning with fine temporal granularity. Additionally, we hope ETRQA can benefit the temporal reasoning research community for future studies.

Recent advancements in large language models (LLMs) have significantly enhanced their ability to understand both natural language and code, driving their use in tasks like natural language-to-code (NL2Code) and code summarisation. However, LLMs are prone to hallucination—outputs that stray from intended meanings. Detecting hallucinations in code summarisation is especially difficult due to the complex interplay between programming and natural languages. We introduce a first-of-its-kind dataset, CodeSumEval, with ~10K samples, curated specifically for hallucination detection in code summarisation. We further propose a novel Entity Tracing Framework (ETF) that a) utilises static program analysis to identify code entities from the program and b) uses LLMs to map and verify these entities and their intents within generated code summaries. Our experimental analysis demonstrates the framework’s effectiveness, leading to a 73% F1 score. The proposed approach provides a method for detecting hallucinations by tracing entities from the summary to the code, allowing us to evaluate summary accuracy and localise the error within the summary.

The rapid adoption of large language models (LLMs) in diverse applications has intensified concerns over their security and integrity, especially in cloud environments where internal model parameters are inaccessible to users. Traditional tamper detection methods, designed for deterministic classification models, fail to address the output randomness and massive parameter spaces characteristic of LLMs. In this paper, we introduce Efficient Sensitive Fingerprinting (ESF), the first fingerprinting method tailored for black-box tamper detection of LLMs. ESF generates fingerprint samples by optimizing output sensitivity at selected detection token positions and leverages Randomness-Set Consistency Checking (RSCC) to accommodate inherent output randomness. Furthermore, a novel Max Coverage Strategy (MCS) is proposed to select an optimal set of fingerprint samples that maximizes joint sensitivity to tampering. Grounded in a rigorous theoretical framework, ESF is both computationally efficient and scalable to large models. Extensive experiments across state-of-the-art LLMs demonstrate that ESF reliably detects tampering, such as fine-tuning, model compression, and backdoor injection, with a detection rate exceeding 99.2% using 5 fingerprint samples, thereby offering a robust solution for securing cloud-based AI systems.

Unsupervised keyphrase prediction has gained growing interest in recent years. However, existing methods typically rely on heuristically defined importance scores, which may lead to inaccurate informativeness estimation. In addition, they lack consideration for time efficiency. To solve these problems, we propose ERU-KG, an unsupervised keyphrase generation (UKG) model that consists of an informativeness and a phraseness module. The former estimates the relevance of keyphrase candidates, while the latter generate those candidates. The informativeness module innovates by learning to model informativeness through references (e.g., queries, citation contexts, and titles) and at the term-level, thereby 1) capturing how the key concepts of documents are perceived in different contexts and 2) estimating informativeness of phrases more efficiently by aggregating term informativeness, removing the need for explicit modeling of the candidates. ERU-KG demonstrates its effectiveness on keyphrase generation benchmarks by outperforming unsupervised baselines and achieving on average 89% of the performance of a supervised model for top 10 predictions. Additionally, to highlight its practical utility, we evaluate the model on text retrieval tasks and show that keyphrases generated by ERU-KG are effective when employed as query and document expansions. Furthermore, inference speed tests reveal that ERU-KG is the fastest among baselines of similar model sizes. Finally, our proposed model can switch between keyphrase generation and extraction by adjusting hyperparameters, catering to diverse application requirements.

Large Language Models (LLMs) have shown impressive reasoning capabilities in well-defined problems with clear solutions, such as mathematics and coding. However, they still struggle with complex real-world scenarios like business negotiations, which require strategic reasoning—an ability to navigate dynamic environments and align long-term goals amidst uncertainty.Existing methods for strategic reasoning face challenges in adaptability, scalability, and transferring strategies to new contexts.To address these issues, we propose explicit policy optimization (*EPO*) for strategic reasoning, featuring an LLM that provides strategies in open-ended action space and can be plugged into arbitrary LLM agents to motivate goal-directed behavior.To improve adaptability and policy transferability, we train the strategic reasoning model via multi-turn reinforcement learning (RL), utilizing process rewards and iterative self-play.Experiments across social and physical domains demonstrate *EPO*’s ability of long-term goal alignment through enhanced strategic reasoning, achieving state-of-the-art performance on social dialogue and web navigation tasks. Our findings reveal various collaborative reasoning mechanisms emergent in *EPO* and its effectiveness in generating novel strategies, underscoring its potential for strategic reasoning in real-world applications. Code and data are available at [https://github.com/lxqpku/EPO](https://github.com/lxqpku/EPO).

Many current studies focus on extracting tests or treatments when constructing clinical pathways, often neglecting the patient’s symptoms and diagnosis, leading to incomplete diagnostic and therapeutic logic. Therefore, this paper aims to extract clinical pathways from electronic medical records that encompass complete diagnostic and therapeutic logic, including temporal information, patient symptoms, diagnosis, and tests or treatments. To achieve this objective, we propose a novel clinical pathway representation: the clinical status pathway. We also design a LLM-based pipeline framework for extracting clinical status pathway from electronic medical records, with the core concept being to improve extraction accuracy by modeling the diagnostic and treatment processes. In our experiments, we apply this framework to construct a comprehensive breast cancer-specific clinical status pathway and evaluate its performance on medical question-answering and decision-support tasks, demonstrating significant improvements over traditional clinical pathways. The code is publicly available at https://github.com/finnchen11/EMRs2CSP.

Recent advancements in Large Language Models (LLMs) show their potential in accurately answering biomedical questions, yet current healthcare benchmarks primarily assess knowledge mastered by medical doctors, neglecting other essential professions. To address this gap, we introduce the Examinations for Medical PErsonnel in Chinese (EMPEC), a comprehensive healthcare knowledge benchmark featuring 157,803 exam questions across 124 subjects and 20 healthcare professions, including underrepresented roles like Optometrists and Audiologists. Each question is tagged for release time and source authenticity. We evaluated 17 LLMs, including proprietary and open-source models, finding that while models like GPT-4 achieved over 75% accuracy, they struggled with specialized fields and alternative medicine. Notably, we find that most medical-specific LLMs underperform their general-purpose counterparts in EMPEC, and incorporating EMPEC’s data in fine-tuning improves performance. In addition, we tested LLMs on questions released after the completion of their training to examine their ability in unseen queries. We also translated the test set into English and simplified Chinese and analyse the impact on different models. Our findings emphasize the need for broader benchmarks to assess LLM applicability in real-world healthcare, and we will provide the dataset and evaluation toolkit for future research.

Electromyography (EMG) tables are crucial for diagnosing muscle and nerve disorders, and advancing the automation of EMG diagnostics is significant for improving medical efficiency. EMG tables contain extensive continuous numerical data, which current Large Language Models (LLMs) often struggle to interpret effectively. To address this issue, we propose EMGLLM, a data-to-text model specifically designed for medical examination tables. EMGLLM employs the EMG Alignment Encoder to simulate the process that doctors compare test values with reference values, aligning the data into word embeddings that reflect health degree. Additionally, we construct ETM, a dataset comprising 17,250 real cases and their corresponding diagnostic results, to support medical data-to-text tasks. Experimental results on ETM demonstrate that EMGLLM outperforms various baseline models in understanding EMG tables and generating high-quality diagnoses, which represents an effective paradigm for automatic diagnosis generation from medical examination table.

Language models today are widely used in education, yet their ability to tailor responses for learners with varied informational needs and knowledge backgrounds remains under-explored. To this end, we introduce ELI-Why, a benchmark of 13.4K “Why” questions to evaluate the pedagogical capabilities of language models. We then conduct two extensive human studies to assess the utility of language model-generated explanatory answers (explanations) on our benchmark, tailored to three distinct educational grades: elementary, high-school and graduate school. In our first study, human raters assume the role of an “educator” to assess model explanations’ fit to different educational grades. We find that GPT-4-generated explanations match their intended educational background only 50% of the time, compared to 79% for lay human-curated explanations. In our second study, human raters assume the role of a learner to assess if an explanation fits their own informational needs. Across all educational backgrounds, users deemed GPT-4-generated explanations 20% less suited on average to their informational needs, when compared to explanations curated by lay people. Additionally, automated evaluation metrics reveal that explanations generated across different language model families for different informational needs remain indistinguishable in their grade-level, limiting their pedagogical effectiveness.

Generative large language models are crucial in natural language processing, but they are vulnerable to backdoor attacks, where subtle triggers compromise their behavior. Although backdoor attacks against LLMs are constantly emerging, existing benchmarks remain limited in terms of sufficient coverage of attack, metric system integrity, backdoor attack alignment. And existing pre-trained backdoor attacks are idealized in practice due to resource access constraints. Therefore we establish ELBA-Bench, a comprehensive and unified framework that allows attackers to inject backdoor through parameter efficient fine-tuning (e.g., LoRA) or without fine-tuning techniques (e.g., In-context-learning). ELBA-Bench provides over 1300 experiments encompassing the implementations of 12 attack methods, 18 datasets, and 12 LLMs. Extensive experiments provide new invaluable findings into the strengths and limitations of various attack strategies. For instance, PEFT attack consistently outperform without fine-tuning approaches in classification tasks while showing strong cross-dataset generalization with optimized triggers boosting robustness; Task-relevant backdoor optimization techniques or attack prompts along with clean and adversarial demonstrations can enhance backdoor attack success while preserving model performance on clean samples. Additionally, we introduce a universal toolbox designed for standardized backdoor attack research at https://github.com/NWPUliuxx/ELBA_Bench, with the goal of propelling further progress in this vital area.

While recent research increasingly emphasizes the value of human-LLM collaboration in competitive programming and proposes numerous empirical methods, a comprehensive understanding remains elusive due to the fragmented nature of existing studies and their use of diverse, application-specific human feedback. Thus, our work serves a three-fold purpose: First, we present the first taxonomy of human feedback consolidating the entire programming process, which promotes fine-grained evaluation. Second, we introduce ELABORATIONSET, a novel programming dataset specifically designed for human-LLM collaboration, meticulously annotated to enable large-scale simulated human feedback and facilitate cost-effective real human interaction studies. Third, we introduce ELABORATION, a novel benchmark to facilitate a thorough assessment of human-LLM competitive programming. With ELABORATION, we pinpoint strengthes and weaknesses of existing methods, thereby setting the foundation for furture improvement. Our dataset and code will be openly released.

Large Language Models (LLMs) have shown remarkable performance in Open-Domain Question Answering (ODQA) by leveraging external documents through Retrieval-Augmented Generation (RAG). To reduce RAG overhead, from longer context, context compression is necessary. However, prior compression methods do not focus on filtering out non-evidential information, which limit the performance in LLM-based RAG. We thus propose Evidentiality-guided RAG, or ECoRAG framework. ECoRAG improves LLM performance by compressing retrieved documents based on evidentiality, ensuring whether answer generation is supported by the correct evidence. As an additional step, ECoRAG reflects whether the compressed content provides sufficient evidence, and if not, retrieves more until sufficient. Experiments show that ECoRAG improves LLM performance on ODQA tasks, outperforming existing compression methods. Furthermore, ECoRAG is highly cost-efficient, as it not only reduces latency but also minimizes token usage by retaining only the necessary information to generate the correct answer. Code is available at https://github.com/ldilab/ECoRAG.

Large Language Models (LLMs) have demonstrated impressive capabilities in language generation and general task performance. However, their application to spoken language understanding (SLU) remains challenging, particularly for token-level tasks, where the autoregressive nature of LLMs often leads to misalignment issues. They also struggle to capture nuanced interrelations in semantic-level tasks through direct fine-tuning alone. To address these challenges, we propose the Entity-level Language Model (ECLM) framework, which reformulates slot-filling as an entity recognition task and introduces a novel concept, Chain of Intent, to enable step-by-step multi-intent recognition. Experimental results show that ECLM significantly outperforms strong baselines such as Uni-MIS, achieving gains of 3.7% on MixATIS and 3.1% on MixSNIPS. Compared to standard supervised fine-tuning of LLMs, ECLM further achieves improvements of 8.5% and 21.2% on these datasets, respectively. Our code is available at https://github.com/SJY8460/ECLM.

Multi-modal Emotion Recognition in Conversation (MMERC) aims to identify speakers’ emotional states using multi-modal conversational data, significant for various domains. MMERC requires addressing emotional causes: contextual factors that influence emotions, alongside emotional evidence directly expressed in the target utterance. Existing methods primarily model general conversational dependencies, such as sequential utterance relationships or inter-speaker dynamics, but fall short in capturing diverse and detailed emotional causes, including emotional contagion, influences from others, and self-referenced or externally introduced events. To address these limitations, we propose the Evidence-Cause Attention Network for Multi-Modal Emotion Recognition in Conversation (ECERC). ECERC integrates emotional evidence with contextual causes through five stages: Evidence Gating extracts and refines emotional evidence across modalities; Cause Encoding captures causes from conversational context; Evidence-Cause Interaction uses attention to integrate evidence with diverse causes, generating rich candidate features for emotion inference; Feature Gating adaptively weights contributions of candidate features; and Emotion Classification classifies emotions. We evaluate ECERC on two widely used benchmark datasets, IEMOCAP and MELD. Experimental results show that ECERC achieves competitive performance in weighted F1-score and accuracy, demonstrating its effectiveness in MMERC

Eligibility criteria (EC) are critical components of clinical trial design, defining the parameters for participant inclusion and exclusion. However, designing EC remains a complex, expertise-intensive process. Traditional approaches to EC generation may fail to produce comprehensive, contextually appropriate criteria. To address these challenges, we introduce EC-RAFT, a method that utilizes Retrieval-Augmented Fine-Tuning (RAFT) to generate structured and cohesive EC directly from clinical trial titles and descriptions. EC-RAFT integrates contextual retrieval, synthesized intermediate reasoning, and fine-tuned language models to produce comprehensive EC sets. To enhance clinical alignment evaluation with referenced criteria, we also propose an LLM-guided evaluation pipeline. Our results demonstrate that our solution, which uses Llama-3.1-8B-Instruct as a base model, achieves a BERTScore of 86.23 and an EC-matched LLM-as-a-Judge score of 1.66 out of 3, outperforming zero-shot Llama-3.1 and Gemini-1.5 by 0.41 and 0.11 points, respectively. On top of that, EC-RAFT also outperforms other fine-tuned versions of Llama-3.1. EC-RAFT was trained in a low-cost setup and, therefore, can be used as a practical solution for EC generation while ensuring quality and relevance in clinical trial design. We release our code on GitHub at https://github.com/biodatlab/ec-raft/

Recent advancements in Large Vision Language Models (LVLMs) show promise for pathological diagnosis, yet their application in clinical settings faces critical challenges of multimodal hallucination and biased responses. While preference alignment methods have proven effective in general domains, acquiring high-quality preference data for pathology remains challenging due to limited expert resources and domain complexity. In this paper, we propose EAGLE (Expert-guided self-enhancement for preference Alignment in patholoGy Large vision-languagE model), a novel framework that systematically integrates medical expertise into preference alignment. EAGLE consists of three key stages: initialization through supervised fine-tuning, self-preference creation leveraging expert prompting and medical entity recognition, and iterative preference following-tuning. The self-preference creation stage uniquely combines expert-verified chosen sampling with expert-guided rejected sampling to generate high-quality preference data, while the iterative tuning process continuously refines both data quality and model performance. Extensive experiments demonstrate that EAGLE significantly outperforms existing pathological LVLMs, effectively reducing hallucination and bias while maintaining pathological accuracy. The source code is available at https://github.com/meidandz/EAGLE.

Mixture-of-Experts (MoE) has demonstrated promising potential in scaling LLMs. However, it is hindered by two critical challenges: (1) substantial GPU memory consumption to load all experts; (2) low activated parameters cannot be equivalently translated into inference acceleration effects. In this work, we propose EAC-MoE, an Expert-Selection Aware Compressor for MoE-LLMs, which deeply aligns with the characteristics of MoE from the perspectives of quantization and pruning, and introduces two modules to address these two challenges respectively: (1) The expert selection bias caused by low-bit quantization is a major factor contributing to the performance degradation in MoE-LLMs. Based on this, we propose Quantization with Expert-Selection Calibration (QESC), which mitigates the expert selection bias by calibrating the routers within the MoE; (2) There are always certain experts that are not crucial for the corresponding tasks, yet causing inference latency. Therefore, we propose Pruning based on Expert-Selection Frequency (PESF), which significantly improves inference speed by pruning less frequently used experts for current task. Extensive experiments demonstrate that our approach significantly reduces memory usage and improves inference speed with minimal performance degradation.

Be it your favorite novel, a newswire article, a cooking recipe or an academic paper – in many daily situations we read the same text more than once. In this work, we ask whether it is possible to automatically determine whether the reader has previously encountered a text based on their eye movement patterns during reading. We introduce two variants of this task and address them using both feature-based and neural models. We further introduce a general strategy for enhancing these models with machine generated simulations of eye movements from a cognitive model. Finally, we present an analysis of model performance which on the one hand yields insights on the information used by the models, and on the other hand leverages predictive modeling as an analytic tool for better characterization of the role of memory in repeated reading. Our work advances the understanding of the extent and manner in which eye movements in reading capture memory effects from prior text exposure, and paves the way for future applications that involve predictive modeling of repeated reading.